Regular announcements: did you know you can hire me for life coaching and general consulting? You can also buy my novella Her Voice Is A Backwards Record wherever fine ebooks are sold (except Google Books).
Effective Altruism
Global Poverty
GiveWell did internal criticism of its own iron grantmaking and found several potential issues: ignoring more recent evidence about the effect of iron supplementation on anemia; potential misestimates of how many people have anemia; not paying enough attention to anemia-related disability in addition to deaths; too much subjectivity in the model; and not enough work in sub-Saharan Africa.
GiveWell plans to consider multiple different data sources to figure out the burden of disease in various areas, rather than relying on a single data source as it did historically. It also plans to more carefully take noise into account.
Particularly Good: I’ve been meaning for years to write a piece explaining what I mean when I say I’m 80% an aid skeptic but the Center for Global Development did it for me. A fantastic read. I highly recommend it to everyone who wants to be less confused about foreign aid.
Animal Advocacy
The number of animals farmed at any time is estimated to double by 2033, with much of the increase driven by invertebrates. By 2033, as many as 75% of farmed animals alive at any time may be invertebrates.
Animal advocates often prioritize animal species based on the number of animals killed, but because animals live for different amounts of time before slaughter, this isn’t a very good way of estimating farmed animal populations. Animal advocates also often don’t consider how intensely an animal suffers or for how long, which can lead to (say) overweighting the importance of slaughter or underweighting the importance of welfare improvements.
Pork lobbyists continue to oppose California’s Proposition 12, which bans selling pork in California if it was produced in a system that confines mother pigs in tiny crates. They want Congress to pass a law forbidding states from refusing to import meat from other states. However, it’s unclear whether Proposition 12 has in fact raised pork prices and whether any increases in prices are temporary results of the adjustment to a new pork-rearing system.
Lessons from history for animal advocates: we’ve been trying to get people to be vegan for a really long time and the number of vegans is basically flat. Even in very vegetarian societies, people can start eating meat. Conversely, changing institutions and laws is an effective tactic. Technology might speed up change.
Predictors of increased animal product consumption include viewing more billboards and activism about animal advocacy (not a typo, more billboards == more meat); consuming fewer plant-based animal product substitutes; and not feeling very motivated to reduce animal product consumption. Predictors of progression towards vegetarianism include participation in No Meat May and Vegan Easy 30; consuming more plant-based animal product substitutes; feeling motivated to reduce animal product consumption; and not handling raw meat much. Also includes an interesting discussion of how collecting data in three waves allows causal inference.
The wild animal welfare movement should prepare for transformative AI. Wild-animal welfare researchers should prioritize research that AIs will have trouble helping with (like data collection) over research AIs will be able to help with (like philosophy and modelbuilding). Wild-animal welfare advocates should prioritize convincing people of the importance of wild animal welfare (especially people who are programming AIs). The spread of wild animals into space is also a key concern. The movement should be flexible and prepared to change its viewpoints.
Actionable: Out of 122 meat alternatives tested in a blind taste test, twenty got similar scores to or outperformed animal meat. The link lists off the brands.
Existential Risk
Using your career to prevent nuclear war. Options include working for a government or international organization, research about nuclear risk, advocacy about nuclear weapons, and nuclear-risk fieldbuilding.
Reasons to be skeptical of AGI by 2030: It’s unclear that we can extrapolate current trends in AI: for example, It’s unclear how well benchmarks generalize to real-world performance. People have been slow to adopt AIs.
But also: “AGI will be here in ten to twenty years” is extremely fast.
What if we analogize AI to outsourcing to India? Communication is important; what you get out of outsourcing depends on how well you manage to communicate with your staff. People don’t like bossing other people around, so AI might be easier to adopt because it feels like ‘going on the computer.’ It’s very hard to create a system that means you reliably hear bad news.
Interpretability isn’t sufficient for AI safety, because it’s easy to miss things, hard to measure progress, and nearly impossible to prove an AI isn’t deceiving you. Nevertheless, interpretability can be part of a broader set of strategies to ensure AIs are safe.
Research priorities for AI welfare: how can we concretely improve AIs’ wellbeing? how can we cooperate with AIs? how can we use AI to advance AI welfare research? how can we figure out whether AIs experience welfare? how can we encourage good conversations about AI welfare?
AI welfare and AI safety share numerous goals, including understanding how AIs think, giving AIs goals which are friendly to humans, and figuring out how to cooperate with AIs.
Natural language processing was the study of getting computers to understand human language. Then large language models were invented. How does a field react to a technological advance making it obsolete?
AI code assistant refuses to write code and tells the human to learn to code instead.
American Democracy
Donald Trump has threatened to ‘unleash the police’, but the police in the United States have never been meaningfully restrained. They can and do violate people’s rights with impunity. They are essentially immune to lawsuits. And for someone who wants the police to be freed from petty concern about civil liberties, Trump sure objects when the police are affecting him.
Meta Effective Altruism
Actionable: The Effective Altruism Infrastructure Fund considers itself funding constrained. I’m excited by the EA Infrastructure Fund’s approach and consider this to be one of the best ways to invest in building the EA community.
The EA Funds are merging with the Centre for Effective Altruism.
I don’t agree with everything in this post—in particular, I approve of 80,000 Hours’s pivot to AI safety—but I found it thought-provoking to read. As effective altruism matures as a movement, it becomes more obvious that ruthless idealism puts some people off and makes it harder to build allyships. But if we become completely pragmatic, we lose the spark that makes the movement special. How do we protect and preserve principles-first EA while still working with people who don’t share our values?
Personal contacts and 80,000 Hours remain the most important ways for people to get into effective altruism. Three people came in through Thing of Things. :D
Other Causes
Unlike the author of this article, I support Section 230, but I was shocked to learn about the role of Facebook in the genocide against the Rohingya. It is fundamentally unacceptable for Facebook to serve 15 million users in Myanmar and have only four content moderators that speak Burmese—especially once Facebook was aware that it was being used as a tool to organize a genocide. I hope any of my readers who work at Meta can work within the company to prevent something like this from happening ever again.
Rationality
Boredom when learning is a signal that you don’t understand the topic.
Politics
As a former New College of Florida student, I’m always eager to read stories about the conservative takeover of my alma mater. The Chronicle of Higher Education has the latest updates. (Athletes! People have to wear shoes! No stickers on mailboxes!) However, no one ever talks about the questions I find most pressing. I want to know how they plan to teach a conservative Great Books program when one of their two classics professors specializes in researching classical obscenity.
Senator John Fetterman is struggling with mental illness. Senior staffers characterize him as ‘manic’ and worry that his mental illness is making him dangerously unstable.
In Arkansas, a Medicaid work requirement led to the removal of 18,000 people from the rolls—but caused no increase in employment among former Medicaid recipients. This is likely because only 3% of Medicaid recipients are working-age adults who are long-term unemployed. Many people I know and love have relied on Medicaid to get medical coverage.
Childcare is important, but there’s no imminent childcare crisis: both labor force participation among women age 25-44 and jobs in the childcare sector continue to rise. The economy is strong, which makes it easier to give women childcare access.
The search for ‘authentic’ writers of color. I was particularly intrigued by the author’s point that diaspora experiences are different from the experiences of people from a culture (sort of by definition), but that the American publishing industry, in its search for ‘authenticity’, has appointed diaspora writers as experts on the cultures their family came from.
Reality Has A Surprising Amount Of Detail
Chinese Doom Scroll: Barnard College is a diploma mill that lets low-achieving women get degrees as prestigious as Columbia’s. Everything on Temu is fake. Makeup work days after holidays. Bus drivers stopping the bus to get lunch or run errands; also, Viagra as a cure for homosexuality. Terminally ill person dies while trying to withdraw money.
Particularly Good: 50 Things I Learned Writing Construction Physics.
Particularly Good: The cargo cults of Papua New Guinea. “imagine for a moment that hyper-advanced aliens have suddenly appeared in your town… when you inquire where their fabulous devices come from and how they work, the aliens explain that, obviously, they perform rituals invoking their ancestors, and then their tribal totems deliver and operate their technology…. Are you going to say “gosh, I guess my entire understanding of the nature of the material universe is wrong, apparently magic is real”? Of course not. You’re not stupid.”
Fiction
Premarin: A very fun piece of Infect Your Friends And Loved Ones fanfiction, about the logistical difficulties in being a badass trans female horse warrior after the apocalypse. (Have you read Infect Your Friends And Loved Ones yet? You should pick up Torrey Peters’ story collection and read Infect Your Friends And Loved Ones. And then read Premarin.)
Born in the wrong generation: So, this is an annoying dunk on LLMs and the author’s political enemies. But it’s also really well-written horror with a delightful ending twist. I think LLMs have been underexplored as a source of the uncanny, , and Kriss does a masterful job here.
Particularly Good: Before you felt her hands: a story about PEPFAR.
Okay, I'm still reading the post on cargo cults, but man this is amazing. It's striking how quickly they came up with new myths to explain strange new phenomena—which fits with suspicions I've had about myths form, but nice to have documented.
Also, this is a hilarious aside: "Heaven, which is not (as you may have been told by some foolish people) in Sydney, Australia, but rather in the sky above Sydney".
Late comment from me (for some reason I was busy this weekend, wink) but:
> Interpretability isn’t sufficient for AI safety, because it’s easy to miss things, hard to measure progress, and nearly impossible to prove an AI isn’t deceiving you.
I continue to believe that *alignment*, as currently defined, is insufficient for AI safety, as I think it analogizes well to making AI instinctively nice. But anything that can truly be called AGI will have the trait, as humans do, that it can choose to do things that go against its instincts. E.g. soldiers train to kill people in a way that allows them to skip past their inmate desire not to.
So even if you make an LLM that won't give me bomb-making instructions instinctively, I don't think it's never gonna give me bomb-making instructions, once it is embedded within some kind of more agentic structure.